taxes, misallocations and productivity · revenue and industrial sales. firms in the same sectors...

Taxes, Misallocations and Productivity

Bruno Caprettini∗

January 5, 2014

Preliminary and Incomplete

Abstract

Misallocations of factors of production have the potential to explain a large

portion of cross-country differences in productivity (Hsieh and Klenow (2009)). Yet,

empirical evidence relating actual differences in firms’ productivity to observable

policy distortions has been scarce. In this paper I exploit a fiscal reform in Brazil

to provide direct empirical evidence of the distortions created by turnover taxes.

Turnover taxes are business taxes that, unlike Value Added Taxes, do not allow to

deduct the cost of intermediate inputs from the tax base, and that for this reason

hit disproportionately industries whose inputs account for a large share of the final

value of production. Using a difference-in-difference approach, I show that after the

reform sectors that rely more on intermediate inputs grow faster in employment,

revenue and industrial sales. Firms in the same sectors that were not affected by

the reform do not show similar patterns of growth during the period. These results

have relevant policy implications, as turnover taxes are very common around the

world.

∗Caprettini: Universitat Pompeu Fabra, Ramon Trias Fargas, 25-27, Barcelona, Spain 08005 (e-mail:[email protected]). I would like to thank my advisors Paula Bustos and Antonio Ciccone forconstant encouragement and support. I also received valuable comments from Ciccio Amodio, AlessandraBonfiglioli, Fernando Broner, Vasco Carvalho, Gino Gancia, Gene Grossman, Eduardo Morales, JacopoPonticelli, Stephen Redding, Esteban Rossi-Hansberg, Alessandro Tarozzi, Joachim Voth, Chris Woodruffand seminar participants at CREI International Lunch, UPF and UAB PhD student seminars and CEPRPEDL Workshop. I carried out part of the project while I was a Visiting Student at the Department ofEconomics at Princeton University: I wish to thank Gene Grossman, Stephen Redding and everybody inthe International Trade Section for their very kind hospitality. Also, I would like to thank Carlos Lessaand Luis-Carlos Pinto at the IBGE for their help to access the firm-level data. Financial support fromthe Private Enterprise Development in Low-Income Countries is also gratefully acknowledged. All errorsare mine.

1

1 Introduction

Productivity differences between rich and poor countries are thought to depend in

large part on misallocations of resources across firms. Policies that alter relative prices

faced by different producers affect how resources are allocated in the economy, and can

prevent productive sectors to expand (Jones (2011b)). Although misallocations have the

potential to explain a large portion of cross-country differences in productivity (Hsieh

and Klenow (2009)) empirical evidence relating actual differences in firms’ productivity

to observable policy distortions has been scarce. The main challenge faced by researchers

is that the number of potential distortions in poor countries is large, and reforms directly

aimed at eliminating some of them are rare. As a result, the size of distortions created

by specific policies is often difficult to assess.

In this paper I exploit a fiscal reform in Brazil to provide direct empirical evidence of

the distortions created by turnover taxes. Turnover taxes are business taxes that are levied

at every stage of production, with no deduction for the cost of intermediate inputs. They

hit disproportionately industries whose inputs account for a large share of the final value

of production, and they create incentives to integrate vertically. Turnover taxes are very

common around the world, especially in low income countries. As of 2013, 60 countries

had some form of tax on turnover in place according to the World Bank, and 22 of them

were classified as low income countries (World Bank (2012)). Although turnover taxes

are generally thought to be very distortive (Diamond and Mirrlees (1971a); Diamond and

Mirrlees (1971b); Jones (2011a); Keen (2013)), there is very little empirical research on

them, and there is virtually no estimate of the magnitude of the distortions that this type

of tax creates.1

I examine how Brazilian manufactures responded to a fiscal reform that eliminated

two different turnover taxes. Between December 2002 and December 2003 the Brazilian

government passed two separate reforms that converted two turnover taxes, PIS and

COFINS, into two value added taxes (VATs), a type of tax that allows to deduct the

cost of intermediate inputs and that for this reason does not hit sectors differentially.

Using a difference-in-difference approach, I compare the different impact of the reform

on sectors that use intermediate inputs intensively and on sectors in which intermediate

inputs account for a lower share of the final value of production. I obtain a sectoral

measure of intermediate input use that is not affected by the reform by taking the United

States as a benchmark, and by assuming that the share going to intermediate inputs in

U.S. sectors is a good proxy for the technical importance of intermediate inputs across

any economy.

1Best et al. (2013) look at turnover taxes in Pakistan, but they document their advantages over a taxon profits in terms of tax enforcement.

2

Using data from the yearly Brazilian industrial survey over the period 1997-2009, I

show that Brazilian sectors that rely more heavily on intermediate inputs grew faster after

the reform in terms of employment, revenues and industrial sales. These results survive

to the inclusion of other sectoral characteristics and they are concentrated only in the set

of firms that are actually affected by the reform. I do not find significant growth in the

sectors with higher intermediate input intensity before the reform, and the fact that I find

faster growth also for employment suggests that changes in the incentives to misreport

revenues or sales can not explain entirely the results. The implied elasticity to the tax is

large, and the results have relevant policy implications, because these gains came at no

expense for tax revenues.

The remainder of the paper is organized as follows. Section 2 provides some back-

ground and explains the 2002-2003 reform. Section 3 introduces the data used and section

4 presents the results. Section 5 concludes.

2 Background

Brazilian manufacturing firms are subject to a complex fiscal regime and have to pay

several different taxes and contributions. This paper focuses on two of the social con-

tributions paid by every manufacturing firm: the Programa de Integracao Social (PIS:

contribution for the social integration programme) and the Contribuicao para Financia-

mento da Seguridade Social (COFINS: contribution for the funding of social security).

Until 2002 these contributions were levied on the total value of turnover, with no de-

ductions for the cost of intermediate inputs.2 The tax rate of PIS was 0.65%.3 When

introduced in 1991, the tax rate of COFINS was 2%4 and the rate was increased to 3%

in 1998.5

Between December 2002 and December 2003 the Brazilian Federal Congress passed

two separate laws that modified the regime of both PIS and COFINS. First, the Law no.

10637 of the 30th December 2002 allowed to deduct the cost of intermediate inputs from

the the tax base of PIS, effectively converting it into a VAT.6 The law also increased the

tax rate of PIS by 1 percentage point, to 1.65%, in order to avoid a fall in fiscal revenues.

Second, on the 29th of December 2003, Congress passed Law no. 10833, that allowed to

2PIS was introduced with the Complementary Law no. 7 on the 7th of September 1970. COFINS wascreated by the Complementary Law no. 70 on the 31st of December 1991; it replaced the existing Fundode Investimento Social (FINSOCIAL) which had been introduced by the Decree no. 1940 of the 25th ofMay 1982 and abolished in 1991 by the same law that introduced COFINS.

3See clause 3 on the Complementary Law no. 7 of the 7th of September 1970 as modified by the clause1 of the Complementary Law no. 17 on the 12th of December 1973.

4See clause 2 of the Complementary Law no. 70 on the 31st of December 1991.5See clause 8 of the Law no. 9718 on the 27th of November of 1998.6See clause 3 of the law. The law converted the Provisionary Measure 66 enacted by the Cardoso

Government on the 29th of August 2002. It came into effect on the 1st of December 2002 (clause 68).

3

deduct the cost of intermediate inputs from the tax base of COFINS.7 The new tax rate

for COFINS was increassed from 3% to 7.6%, a number that was chosen by applying the

same multiplier adopted to set the new rate of PIS under the VAT regime (roughly 2.53;

see Werneck (2006) for a discussion about the debate that led to this choice).

The laws excluded from the new system a number of firms that continued paying

PIS and COFINS as turnover taxes at lower rates. The most relevant of these exception

concerns small firms that opt to pay taxes under a simplified tax regime called SIMPLES.8

These firms were not allowed to adopt the new PIS and COFINS and still today pay them

on the total value of their turnover at a rate of 0.65% and 3%. In my analysis I focus on

firms that do not opt for SIMPLES and thus are affected by the reform.

The reforms were highly visible and publicized even to the general public. Moreover,

firms pay both PIS and COFINS monthly, and they had incentives to adapt to the new

system quickly after the reform. This may help explain the reason of the speed with which

some of the outcomes reacted to the reform.

The stated objective of the reforms was to increase competitiveness of the Brazil-

ian manufacturing sector, and the policies were strongly encouraged by the International

Monetary Fund (IMF) as part of the agreements reached with the international organi-

zation to obtain recovery loans in the aftermath of the capital account crisis of 1998.9

The reform was received with criticism by representatives of agriculture and the service

sectors, who were worried that the savings generated by lower intermediate input taxation

would be offset by the increase in the tax rate for firms that operated in in sectors where

intermediate inputs do not account for a large share of the final value of production. In

what follows I focus on manufacturing firms, which arguably stood to gain most from the

reform.10

Finally, PIS and COFINS reforms can be analyzed together because the two laws were

passed in a relatively short period. Moreover, the structure of PIS and COFINS was very

similar before 2002, and the laws that modified them in 2002 and 2003 were similar in

content, structure and even wording. Finally, the first of the two laws, beside modifying

PIS, also contained the express provision that within one year the Government should

present to the Congress a law proposal to turn COFINS into a VAT.11 Thus, firms and

7See clause 3 of the law. The law converted the Provisionary Measure 135 enacted by the LulaGovernment on the 30th of October 2003. It came into effect on the 2nd of February 2004 (clause 93).

8SIMPLES was introduced on the 5th of December 1996 by the Law 9317. The system allows firmsthat report revenues below 720 thousand Brazilian Reais to opt for a simplified tax regime where mosttaxes are paid together, and tax rates are lower. The threshold below which firms can opt for theSIMPLES was raised to 1,2 millions Brazilian Reais by the law 9732 of the 11th of December 1998 andto 2,4 millions Brazilian Reais by the Law no. 11196, of the 21st of November 2005.

9See press releases IMF (2002) and IMF (2003), for instance.10Note also that most of the sectors that could have lost due to the reform successfully lobbied and

obtained either a reduction in the tax rate or the possibility to remain under the old regime.11See clause 12 in the Law no. 10637 of the 30th December 2002.

4

the general public were expecting a change in COFINS already in 2002 when PIS was

modified.

3 Data

I source firm and plant-level data from the Pesquisa Industrial Anual, (PIA), the

Brazilian yearly industrial survey carried out by the Instituto Brasileiro de Geografia e

Estatıstica (IBGE), the Brazilian National Statistical Office.

The survey contains yearly information for the years 1997 through 2009.12 It covers

formal firms operating in the manufacturing sectors, recorded under the Classificacao

Nacional de Atividades Economicas (CNAE) codes 15 through 37 (corresponding to the

ISIC, Rev.3 categories C and D).13 Firms sampled in PIA are drawn from two strata.

First, manufacturing firms employing 30 or more employees are sampled with probability

1 every year. Second, every year IBGE interviews a random sample of manufacturing firms

employing between 5 and 29 workers. The sample is designed in order to be representative

of the Brazilian manufacturing sector within Federal Units and 3-digits CNAE sectors, and

for each firm in this second stratum IBGE computes expansion weights.14 The database

is an unbalanced panel, since anonimized fiscal identifiers allow to track firms from one

year to another.

PIA is made of two parts. The first one, has information on firms, while the second

one covers establishments. Both parts are designed after the U.S. Census of Manufactures

(CM) and the Annual Survey of Manufactures (ASM). At the firm level, IBGE uses two

different questionnaires for the two strata: firms that are sampled with probability 1

answer to a questionnaire that is more thourough than the questionnaire that is presented

to firms that are sampled only probabilistically. Basic information on revenues, costs,

employement, investments and whether or not the firm opted to pay taxes under the

SIMPLES regime is available for firms in both strata.15 Every firm that operates more

12The survey is available since 1996, but I exclude the first year because the SIMPLES regime wasintroduced only in 1997. As explained in section 2, the PIS and COFINS reform did not affect the firmsthat opted for the simplified regime SIMPLES. For this reason, I show all my results only looking atthe group of affected firms, and drop the year 1996 where I can not identify firms that would opt forSIMPLES.

13CNAE classification changed twice during the years under study. In 2003, with the introduction ofthe ISIC Rev. 3.1, CNAE 1.0 replaced the original CNAE classification. In 2007, with the fourth revisionof the ISIC, CNAE 2.0 replaced the CNAE 1.0. IBGE classifies all firms in PIA according to the currentclassification, and in order to construct a time consistent sector indicator I build a bridge across thethree version of CNAE following the concordances suggested by the Comissao Nacional de Classificacao(CONCLA), the Brazilian National Commission of Classifications.

14Every year, IBGE constructs the two strata of firms by using the administrative records kept bythe Ministry of Labor in the Relacao Anual de Informacoes Sociais (RAIS), the yearly report of socialinformation. The report is carried out yearly and covers the population of Brazilian formal employees(IBGE (2008)).

15Firms sampled with probability 1 report more disaggregated information on investments, work in

5

than one establishment reports information both for the whole firm and for each of the

establishments that it owns. At the establishment level, PIA records information on the

municipality where the plant is located, the revenues, costs and employment.

The PIA database contains a total of 483222 firm-year observations from 107790 unique

firms. Of these, 364454 observations (or 75%) are from firms that are sampled with

probability 1 and completed the longer questionnaire. Although these large firms account

for around 21% of the total number of Brazilian manufacturing firms with 5 or more

employees, they employ around 80% of the total manufacturing workers, and account for

around 95% of total value of shipments.16 The establishment database contains 242165

establishment-year observations, from 53685 unique establishments.17 Table 1 presents

descriptive statistics for both firms and plants.

4 Empirics

4.1 Baseline results

In this section I document how the conversion of PIS and COFINS into VATs led

to a reallocation of production shares towards sectors that use intermediate inputs more

intensively.

Turnover taxes distort the allocation of resources across sectors because, by taxing

intermediate inputs, they effectively levy a higher tax rate on sectors that rely more on

them. As a result, sectors that use relatively more intermediate inputs per unit of output

grow less than they would absent the turnover tax. Since a VAT introduces no such

distortion, the conversion of PIS and COFINS into VATs should lead sectors that use

intermediate inputs more intensively to grow faster after the 2002-2003 reforms.

In order to test this prediction I adopt a difference-in-difference approach, and test

whether sectors that are more intermediate intensive grow faster after the reform com-

pared to sectors that are less intermediate intensive. I obtain a measure of “intermediate-

intensity” that is exogenous to the structure of the Brazilian economy and to the 2002-2003

reform by using the United States as a benchmark. In particular, for every U.S. manufac-

turing sector, I use the NBER Productivity database (Bartelsman and Gray (1996)) to

compute a measure of intermediate input intensity. This is constructed as the expendi-

progress and trade participation.16Informal employment is very common in Brazil and PIA entirely misses it. The 2000 Brazilian

Population Census records almost 8.8 millions people employed in manufacturing, which means that the5 millions people covered by PIA in that year account for around 58% of the total employment in thesector.

17Of these, only 178637 observations are manufacturing establishments (38875 unique establishments):the others are commercial outlets of manufacturing firms.

6

ture on all materials except fuels and electricity divided by the total value of shipment.18

Table 2 reports intermediate input intensity for all U.S. 4-digits manufacturing sectors.

I use this measure of intermediate intensity to explain the performance of Brazilian

sectors with the following difference-in-difference specification:

yst = αs + αt + β · Post2002t × Intermediate Intensitys + est (1)

In (1) subscripts identify years (t) and 4-digits CNAE sectors (s). As dependent

variable yst I consider different measures of industrial performance (described shortly).

On the right-hand side, αt and αs are year and sector fixed effects; Post2002 is a dummy

equal to 1 every year between 2003 and 2009 and Intermediate Intensity is the share

going to intermediate inputs in U.S. sectors described above. The coefficient of interest

is β and I expect sectors that are more intermediate intensive to grow faster after the

reform: β > 0.

On the left-hand side, I regress 4 different outcomes: the total number of workers

employed in establishments operating in year t and sector s, the total value of revenues

collected by establishments operating in year t and sector s, the total value of industrial

goods sold by establishments operating in year t and sector s and the total number of

establishments operating in year t and sector s.19 Since the reform affected only firms

that did not pay taxes according to the simplified tax regime SIMPLES I construct the

four outcomes by using the information from firms that did not opt for this regime. These

represent around 30% of Brazilian firms with at least 5 employees, but account for 97%

of revenues and 75% of employment.

Consistent identification of β requires that sectors with different intermediate intensity

were not following different paths before the reform (common trends assumption) and that

the composition of sectors did not change differentially after the reform (this requires for

instance that the fastest growing firms were not more likely to enter a specific sector

18More precisely, the measure of intermediate input intensity is calculated for every sector as theaverage between 1996 and 2002 of the total expenditure on materials (matcost) minus the value of fueland electricity (energy), divided by the total value of shipments (vship). The NBER Productivity databaseis classified according to the NAICS 1997 sector classification, while plants and firms in the PIA databaseare classified according to the Brazilian CNAE, CNAE 1.0 or CNAE 2.0, depending on the year. I use thecorrespondence between NAICS 1997 and CNAE proposed by Muendler (2002) and the correspondencesbetween CNAE and CNAE 1.0 and CNAE 1.0 and CNAE 2.0 proposed by the Brazilian CONCLA tobridge US sectors to Brazilian sectors between 1997 and 2009.

19In principle I would like to use only the total value of industrial sales, as turnover taxes affectsprimarily industrial production, while revenues include resales and revenues from business services. Un-fortunately, PIA reports either industrial sales gross of taxes or after tax revenues. Since the 2002-2003tax reform that converted PIS and COFINS into VATs increased the tax rate of both taxes, it is im-portant to use a measure of sales that does not contain taxes neither before nor after the reform. Forthis reason I estimate industrial sales net of taxes following the procedure suggested by IBGE (2010) andreport results both with this estimated variable and with the less correct (but directly observed) aftertax revenue.

7

in response to the reform). Given the relatively long time series I am going to be able

to present results that suggest that the common trends assumption is warranted in this

setup.

Table 3 presents the estimates of equation (1). The 4 columns of the table report the

estimates of β for the four outcomes analyzed: employment, industrial sales, revenues and

number of firms. These coefficients are all positive and significant at the 1% level with

the exception of the coefficient of the number of firms. The magnitude of the effect is

large. The interquartile range of the intermediate intensity measure is 0.09, thus the point

estimates in table 3 imply that, relative to the sector in the 25th percentile (Manufacture

of spirits), the sector on the 75th percentile (Manufacture of knitted clothing) expanded

employment by 6.12% more, revenues by 8.78%, industrial sales by 9.48% and the number

of operating establishments by 3.14% (not significant). During the same period, the

average manufacturing sector grew by 2.71% in employment, 4.13% in revenues and 4.14%

in industrial sales.

4.2 Robustness

In this section I show that the results presented in table 3 are robust to a wide range

of tests.

A first concern with the findings shown in section 4.1 is that the effect is entirely

driven by a change in misreporting behavior of Brazilian firms induced by the reform.

Pomeranz (2013) has shown that value added taxes facilitate tax enforcement by creating

a paper trail on transactions between firms. Since the 2002-2003 Brazilian fiscal reform

converted PIS and COFINS into VATs, it is possible that it also affected the incentives

of Brazilian firms to misreport their revenues and sales. If firms operating in sectors

that use intermediate inputs more intensively had greater incentives to misreport their

activity with a turnover tax than with a value added tax, then the effect shown in table

3 would be driven by a change in misreporting behavior and not by real changes in the

economy. Although it is not possible to exclude that the reform affected differentially the

incentives to misreport, table 3 shows that the reform had an effect also on employment.

Since the reform did not change the incentives to report employment truthfully, this result

suggests that changes in the incentives to misreport revenues or sales can not explain all

the findings, and that the reform had some real effects too.

A second concern with the results shown in table 3 is that the intensity of intermediate

input use is correlated with other characteristics of sectors that affected their performance

differentially. First, sectors that are more intensive in the use of capital may have benefited

more of the quieter financial atmosphere during the recovery of the 1998 capital account

crisis. Capital intensity and intermediate intensity are correlated (correlation = 31%,

8

significant < 0.01%) and it is possible that the measure of intermediate intensity is picking

up part of this variation after 2002. Also skill intensity is significantly correlated with

intermediate intensity (correlation = −29%, significant < 0.01%) and, although it is not

obvious why less skill intensive sector should grow less after 2002, one may still worry

that the measure of intermediate intensity picks up some of the variation generated by

skill intensity. In order to control for these possible omitted variable I augment regression

(1) in the following way:

yst = αs + αt + β · Post2002t × Intermediate Intensitys +

+γ · Post2002t ×K Intensitys + δ · Post2002t × Skill Intensitys + est (2)

I compute both capital intensity and skill intensity using information in the NBER

Productivity database and following the procedure proposed by Nunn (2007).20 Table

4 reports the results of (2) for all outcomes including capital and skill intensity. All

results shown in the previous section go through: the coefficients of interest remain sig-

nificant, the point estimate increases in size slightly and the β on the number of firms

becomes significant at the 10% level. In contrast, neither of the coefficients on the two

new interactions is significant in any specification.

A third concern is that sectors that use relatively more intermediate inputs may have

grown faster everywhere in the world, maybe driven by innovation or world demand. If

the results shown in table 3 were replicated for the sectors in the United States, then

we would worry that the effect documented above is not driven by the Brazilian tax

reform but rather by some trend that is common around the world. In order to dispel this

concern, I run regression (1) using the U.S. data on value of workers, shipments, and value

added from the NBER Productivity database. I focus on the same period on which I ran

regression (1) for Brazil and look only at the years between 1997 and 2009. Notice that the

data are very comparable because the Brazilian PIA is designed after the US ASM survey,

and the NBER Productivity database is constructed using yearly information from this

source (Bartelsman and Gray (1996)). Table 5 reports the results for the 3 regressions

run on US data. The table shows that sectors that relied on intermediate inputs did not

experience faster growth in the US after 2002.

A fourth concern is that faster growth of sectors that use intermediate inputs more

intensively is driven by a demand trend that is unrelated to the tax reform. For instance

it is possible that sectors that use intermediate inputs more intensively are also sectors

20More precisely, I compute capital intensity as the natural logarithm of the real value of capital dividedby value added between 1996 and 2002: cap / vadd. I compute skill intensity as 1 minus the percentageof wages to production worker in total payroll between 1996 and 2002: 1 - (prodw / pay). Note also thatthe results shown in table 4 are robust to adopting the definition of skill and capital intensity proposedby Levchenko (2007) instead (results available upon request).

9

that produce goods with higher income elasticity. Since real income has grown steadily in

Brazil during the last decade, this may be driving the results shown in table 3. However,

if a trend that is common to the whole Brazil is truly driving these results, then the

effects documented in table 3 should be common to all the firms operating in Brazil, not

only the ones affected by the reform. Since firms that opted for the SIMPLES were not

affected by the reform, I can use the information coming from them to perform a simple

test of whether firms operating in intermediate intensive sectors grew relatively faster

after the reform even when they were not interested by the reform. Table 6 reports the

test. It shows the coefficients of regression (1) when the outcome yst are computed using

the information from firms that opted to pay taxes according to the simplified tax regime

SIMPLES. In this case all coefficients are statistically equal to 0. These results should

be taken with care, because the firms that opted for SIMPLES are systematically smaller

than the firms that did not opt for the SIMPLES, and thus they can not be considered a

valid counterfactual for them. Nevertheless, the test is informative, because it allows to

rule out the possibility that the results in table 3 are driven by an underlying trend that

is common to the whole Brazil.

Finally, as mentioned in the previous section, the β reported in table 3 would not

be a consistent estimate of the effect of the reform if sectors that use relatively more

intermediate inputs were growing faster than other sectors already before the reform (the

common trend assumption). PIA covers 6 full years before the reform (1997 through 2002),

and thus it allows to perform a suggestive test of the common trends assumption. Table

7 reports the test. It shows the results of running regression (1) using only data between

1997 and 2002 and by assuming that a “placebo” tax reform happened in 1999. The

coefficient reported in the table are thus the coefficient of the interaction Post1999t×Intermediate Intensitys, where Post1999t is a dummy variable equal to 1 for all years

between 2000 and 2002. As table 7 shows, no significant growth can be found in sectors

that use intermediate inputs more intensively before 2002. However, a possible critique

to this test is that it is performed on a sample of sectors-years that is less than half of

the sample used for the results reported in table 3. A more valid comparison would be

with a regression performed over exactly 6 years that straddle the year of the reform.

Table 8 shows the results of such a regression. It reports the coefficient of regression (1)

using only data for the years 2000 through 2005 and the “correct” date for the tax reform

(i.e. Post2002t). Table 8 makes it clear that the insignificant effect estimated before 2002

can not be explained by the sample size, as regressions performed with roughly the same

sample size in the correct period are all significant at the 5% level.

10

5 Conclusions

Misallocations of factors of production have the potential to explain a large portion

of the cross-country differences in productivity (Hsieh and Klenow (2009), Restuccia and

Rogerson (2008)). Yet, despite the potential importance of misallocations in explaining

differences in development around the world, there is still scarce empirical evidence re-

lating existing productivity differences to observable policy distortions. In this paper I

exploit a fiscal reform that happened in Brazil between 2002 and 2003 to provide direct

empirical evidence of the distortions introduced by some types of taxes. In particular, I

focus on turnover taxes, a type of business tax that is levied on the full value of production

rather than on the value added, and that for this reason hits more heavily those sectors

in which intermediate inputs account for a larger share of the final value of production.

Using a difference-in-difference approach, I show that after the reform sectors that

rely more on intermediate inputs grow faster in employment, revenues and industrial

sales. The effect is not present when I look at a set of firms that was not affected by

the reform, and does not appear to be driven by omitted sectoral characteristics nor

existing pre-trends. To the extent that the new allocation of resources better reflects

the technological possibilities of the Brazilian economy as well as the tastes of Brazilian

consumers, the reform brought net welfare gains, because it did not reduce tax revenues.

These results provide the first empirical quantification of the distortions introduced by

turnover taxes, and they have relevant policy implication as these taxes are very common

around the world.

11

References

Bartelsman, E. J. and W. B. Gray (1996). NBER-CES manufacturing industry database.

NBER Technical Working Paper 205.

Best, M. C., A. Brockmeyer, H. J. Kleven, J. Spinnewijn, and M. Waseem (2013). Pro-

duction vs revenue efficiency with limited tax capacity: Theory and evidence from

pakistan.

Diamond, P. A. and J. A. Mirrlees (1971a). Optimal taxation and public production I:

Production efficiency. The American Economic Review 61 (1), 8–27.

Diamond, P. A. and J. A. Mirrlees (1971b). Optimal taxation and public production II:

Tax rules. The American Economic Review 61 (3), 261–278.

Hsieh, C.-T. and P. J. Klenow (2009). Misallocation and manufacturing TFP in china

and india. The Quarterly Journal of Economics 124 (4), 1403–1448.

IBGE (2008). Estatısticas do Cadastro Central de Empresas 2006. Rio de Janeiro, Brazil:

Instituto Brasileiro de Geografia e Estatıstica - IBGE.

IBGE (2010). Pesquisa Industrial Anual — Empresa, Volume 26 of Serie Relatorios

Metodologicos. Rio de Janeiro, Brazil: Instituto Brasileiro de Geografia e Estatıstica -

IBGE.

IMF (2002). IMF approves us$ 30.4 billion stand-by credit for Brazil. Press Release 02\40.

IMF (2003). IMF approves 15-month extension, and us$6.6 billion augmentation of

Brazil’s stand-by credit. Press Release 03\217.

Jones, C. I. (2011a). Intermediate goods and weak links in the theory of economic devel-

opment. American Economic Journal: Macroeconomics 3 (2), 1–28.

Jones, C. I. (2011b). Misallocation, economic growth, and input-output economics. NBER

Working Paper Series 16742.

Keen, M. (2013). Targeting, Cascading, and Indirect Tax Design. International Monetary

Fund.

Levchenko, A. A. (2007). Institutional quality and international trade. The Review of

Economic Studies 74 (3), 791–819.

Muendler, M.-A. (2002). Definitions of Brazilian mining and manufacturing sectors and

their conversion. Mimeo.

12

Nunn, N. (2007). Relationship-specificity, incomplete contracts, and the pattern of trade.

The Quarterly Journal of Economics 122 (2), 569–600.

Pomeranz, D. D. (2013). No taxation without information: Deterrence and self-

enforcement in the value added tax.

Restuccia, D. and R. Rogerson (2008). Policy distortions and aggregate productivity with

heterogeneous establishments. Review of Economic Dynamics 11 (4), 707–720.

Werneck, R. L. F. (2006). An evaluation of the 2003 tax reform effort in Brazil. Revista

de Economia Polıtica 26 (1), 75–94.

World Bank (2012). Doing business 2013: smarter regulations for small and medium-size

enterprises. World Bank Publications.

13

Tables

Table 1Summary Statistics

Firms Multi-plant firmsSIMPLES Non-SIMPLES Firms Plants

Firm-years 205’655 269’493 Observations 46’336 180’584

Workers WorkersMean 35 200 Mean 556 111Median 28 66 Median 129 18St. Dev. 54 796 St. Dev. 1774 413

Revenues (million 2000 R$) Revenues (million 2000 R$)Mean 0.56 27.55 Mean 94.61 25.53Median 0.30 3.40 Median 8.39 0.78St. Dev. 10.13 378.66 St. Dev. 892.14 217.55

Note: Source: Pesquisa Industrial Anual (PIA), years 1997-2009 (IBGE (2010)). The first two columns of

the table reports summary statistics for number of observations, employment and revenues for all firms in

the survey. The first column reports statistics for firms that opted for the simplified tax regime (SIMPLES)

and the second column for firms that did not opt for it. The third column reports summary statistics for

all firms that report more than one non-administrative establishment (multi-plant firms). The last column

reports summary statistics for each establishment owned by a multi-plant firm.

14

Table 2Intermediate intensity of 4-digits NAICS sectors

NAICS NAICS definition Intermediate NAICS NAICS definition Intermediatecode Intensity code Intensity

3241 Petroleum and Coal Products Manuf. 0.751 3359 Other Electrical Equip. and Component Manuf. 0.4723361 Motor Vehicle Manuf. 0.712 3321 Forging and Stamping 0.4693343 Audio and Video Equip. Manuf. 0.671 3332 Industrial Machinery Manuf. 0.4683211 Sawmills and Wood Preservation 0.667 3353 Electrical Equip. Manuf. 0.4643115 Dairy Product Manuf. 0.658 3114 Fruit and Vegetable Preserving and Specialty Food Manuf. 0.4633116 Animal Slaughtering and Processing 0.648 3251 Basic Chemical Manuf. 0.4623362 Motor Vehicle Body and Trailer Manuf. 0.641 3261 Plastics Product Manuf. 0.4623131 Fiber, Yarn, and Thread Mills 0.632 3259 Other Chemical Product and Preparation Manuf. 0.4593117 Seafood Product Preparation and Packaging 0.621 3262 Rubber Product Manuf. 0.4573111 Animal Food Manuf. 0.610 3371 Household and Institutional Furniture and Kitchen Cabinet Manuf. 0.4553161 Leather and Hide Tanning and Finishing 0.610 3342 Communications Equip. Manuf. 0.4543314 Nonferrous Metal (except Aluminum) Production and Processing 0.608 3366 Ship and Boat Building 0.4493312 Steel Product Manuf. from Purchased Steel 0.606 3326 Spring and Wire Product Manuf. 0.4413313 Alumina and Aluminum Production and Processing 0.590 3399 Other Miscellaneous Manuf. 0.4373141 Textile Furnishings Mills 0.586 3253 Pesticide, Fertilizer, and Other Agricultural Chemical Manuf. 0.4373341 Computer and Peripheral Equip. Manuf. 0.578 3333 Commercial and Service Industry Machinery Manuf. 0.4343363 Motor Vehicle Parts Manuf. 0.569 3325 Hardware Manuf. 0.4313212 Veneer, Plywood, and Engineered Wood Product Manuf. 0.566 3113 Sugar and Confectionery Product Manuf. 0.4283365 Railroad Rolling Stock Manuf. 0.563 3351 Electric Lighting Equip. Manuf. 0.4253219 Other Wood Product Manuf. 0.557 3273 Cement and Concrete Product Manuf. 0.4243369 Other Transportation Equip. Manuf. 0.556 3221 Pulp, Paper, and Paperboard Mills 0.4173331 Agriculture, Construction, and Mining Machinery Manuf. 0.556 3346 Manuf. and Reproducing Magnetic and Optical Media 0.4133324 Boiler, Tank, and Shipping Container Manuf. 0.549 3119 Other Food Manuf. 0.4103252 Resin, Synth. Rubber, & Artificial Synth. Fibers & Filaments Manuf. 0.548 3329 Other Fabricated Metal Product Manuf. 0.4083352 Household Appliance Manuf. 0.542 3169 Other Leather and Allied Product Manuf. 0.4033222 Converted Paper Product Manuf. 0.536 3372 Office Furniture (including Fixtures) Manuf. 0.3983133 Textile and Fabric Finishing and Fabric Coating Mills 0.530 3274 Lime and Gypsum Product Manuf. 0.3923311 Iron and Steel Mills and Ferroalloy Manuf. 0.526 3231 Printing and Related Support Activities 0.3823132 Fabric Mills 0.523 3315 Foundries 0.3743255 Paint, Coating, and Adhesive Manuf. 0.519 3328 Coating, Engraving, Heat Treating, and Allied Activities 0.3723149 Other Textile Product Mills 0.510 3256 Soap, Cleaning Compound, and Toilet Preparation Manuf. 0.3703151 Apparel Knitting Mills 0.507 3279 Other Nonmetallic Mineral Product Manuf. 0.3593334 Ventilation, Heating, Air-Conditioning, & Comm. Refrigeration Manuf. 0.506 3345 Navigational, Measuring, Electromedical, & Control Instr. Manuf. 0.3513336 Engine, Turbine, and Power Transmission Equip. Manuf. 0.505 3335 Metalworking Machinery Manuf. 0.3463162 Footwear Manuf. 0.499 3322 Cutlery and Handtool Manuf. 0.3443379 Other Furniture Related Product Manuf. 0.497 3118 Bakeries and Tortilla Manuf. 0.3443112 Grain and Oilseed Milling 0.493 3272 Glass and Glass Product Manuf. 0.3363152 Cut and Sew Apparel Manuf. 0.493 3327 Machine Shops; Turned Product; and Screw, Nut, and Bolt Manuf. 0.3323323 Architectural and Structural Metals Manuf. 0.491 3391 Medical Equip. and Supplies Manuf. 0.3053121 Beverage Manuf. 0.489 3344 Semiconductor and Other Electronic Component Manuf. 0.3043339 Other General Purpose Machinery Manuf. 0.485 3254 Pharmaceutical and Medicine Manuf. 0.2993159 Apparel Accessories and Other Apparel Manuf. 0.482 3271 Clay Product and Refractory Manuf. 0.2863364 Aerospace Product and Parts Manuf. 0.475 3122 Tobacco Manuf. 0.151

Note: Source: NBER Productivity database (Bartelsman and Gray (1996)). Intermediate intensity is calculated for every sector as the average between 1996 and 2002 of the total expenditureon materials (matcost) minus the value of fuel and electricity (energy), divided by the total value of shipments (vship).

15

Table 3The effect of the reform on sectoral allocation of production

Employment Revenues Industrial Sales Firms

Post2002×Intermediate Intensity 0.654*** 0.976*** 1.013*** 0.349

(0.235) (0.285) (0.281) (0.243)

Sector FE (245 sectors) Yes Yes Yes YesYear FE (13 years) Yes Yes Yes YesObservations 2,978 2,978 2,978 2,978R-squared 0.951 0.943 0.943 0.939

Note: The table reports OLS estimates of the coefficient β in equation (1) in the text. The units

of observation are 4-digits CNAE sector-year. Employment is the natural logarithm of the total

number of workers employed in establishments operating in year t and sector s. Revenues is the

natural logarithm of the total value of after tax revenues collected by establishments operating in

year t and sector s. Industrial sales is the natural logarithm of the after tax value of industrial goods

sold by establishments operating in year t and sector s, where the after tax value of sales is estimated

using the procedure suggested by IBGE (2010). Number of firms is the natural logarithm of the total

number of establishments operating in year t and sector s. Intermediate intensity is calculated for

every sector as the average between 1996 and 2002 of the total expenditure on materials (matcost)

minus the value of fuel and electricity (energy), divided by the total value of shipments (vship).

Post2002 is a dummy variable equal to 1 every year after 2002, and 0 otherwise. The source

of dependent variables is Pesquisa Industrial Anual (PIA), years 1997-2009 (IBGE (2010)). The

source of intermediate intensity is the NBER Productivity database (Bartelsman and Gray (1996)).

Robust standard errors clustered at sector level are reported in parentheses. Significance levels:

*** p<0.01, ** p<0.05, * p<0.1.

16

Table 4The effect of the reform on sectoral allocation of productionRobustness of results reported in table 3 to the inclusion of sectoral character-

istics


Post2002×Intermediate Intensity 0.852*** 0.968*** 0.999*** 0.488*

(0.273) (0.349) (0.347) (0.250)

Skill Intensity 0.130 -0.136 -0.156 0.146(0.243) (0.266) (0.266) (0.212)

Capital Intensity -0.095 -0.024 -0.025 -0.055(0.068) (0.107) (0.107) (0.050)


Note: The table reports OLS estimates of the coefficient β in equation (2) in the text. The units

of observation are 4-digits CNAE sector-year. Employment is the natural logarithm of the total

number of workers employed in establishments operating in year t and sector s. Revenues is the

natural logarithm of the total value of after tax revenues collected by establishments operating in

year t and sector s. Industrial sales is the natural logarithm of the after tax value of industrial

goods sold by establishments operating in year t and sector s, where the after tax value of sales is

estimated using the procedure suggested by IBGE (2010). Number of firms is the natural logarithm

of the total number of establishments operating in year t and sector s. Intermediate intensity is

calculated for every sector as the average between 1996 and 2002 of the total expenditure on

materials (matcost) minus the value of fuel and electricity (energy), divided by the total value of

shipments (vship). Capital intensity is the natural logarithm of the average real value of capital

divided by value added between 1996 and 2002: cap / vadd. Skill intensity is the average of 1

minus the percentage of wages to production worker in total payroll between 1996 and 2002: 1 -

(prodw / pay). Post2002 is a dummy variable equal to 1 every year after 2002, and 0 otherwise.

The source of dependent variables is Pesquisa Industrial Anual (PIA), years 1997-2009 (IBGE

(2010)). The source of the intermediate intensity, capital intensity and skill intensity is the NBER

Productivity database (Bartelsman and Gray (1996)). Robust standard errors clustered at sector

level are reported in parentheses. Significance levels: *** p<0.01, ** p<0.05, * p<0.1.

17

Table 5The effect of the reform on sectoral allocation of pro-ductionFalsification test of results reported in table 3: checking for the effect

on U.S. sectors

Employment Shipments Value Added

Post2002×Intermediate Intensity -0.059 0.014 0.149

(0.127) (0.154) (0.163)

Sector FE (473 sectors) Yes Yes YesYear FE (13 years) Yes Yes YesObservations 6,149 6,149 6,149R-squared 0.960 0.953 0.946

Note: The table reports OLS estimates of the coefficient β in equation (1) in the

text when the dependent variables are constructed using U.S. data. The units of

observation are 6-digits NAICS sector-year. Employment is the natural logarithm of

the total number of employees working in year t and U.S. sector s (emp). Shipments is

the natural logarithm of the total value of industry shipments in year t and U.S. sector

s (vship). Value added is the natural logarithm of the value added by manufactures

in year t and U.S. sector s. Intermediate intensity is calculated for every sector as the

average between 1996 and 2002 of the total expenditure on materials (matcost) minus

the value of fuel and electricity (energy), divided by the total value of shipments (vship).

Post2002 is a dummy variable equal to 1 every year after 2002, and 0 otherwise. The

source of dependent variables and intermediate intensity is the NBER Productivity

database (Bartelsman and Gray (1996)). Robust standard errors clustered at sector

level are reported in parentheses. Significance levels: *** p<0.01, ** p<0.05, * p<0.1.

18

Table 6The effect of the reform on sectoral allocation of productionFalsification test of results reported in table 3: checking for the effect on firms

opting for the SIMPLES regime


Post2002×Intermediate Intensity 0.364 -0.224 -0.209 0.374

(0.564) (0.745) (0.764) (0.474)


Note: The table reports OLS estimates of the coefficient β in equation (1) in the text when the

dependent variables are constructed using information from firms that did not opt for the simplified

tax regime SIMPLES. The units of observation are 4-digits CNAE sector-year. Employment is the

natural logarithm of the total number of workers employed in establishments operating in year t

and sector s that did not opt for SIMPLES. Revenues is the natural logarithm of the total value

of after tax revenues collected by establishments operating in year t and sector s that did not opt

for SIMPLES. Industrial sales is the natural logarithm of the after tax value of industrial goods

sold by establishments operating in year t and sector s that did not opt for SIMPLES, where the

after tax value of sales is estimated using the procedure suggested by IBGE (2010). Number of

firms is the natural logarithm of the total number of establishments operating in year t and sector

s that did not opt for SIMPLES. Intermediate intensity is calculated for every sector as the average

between 1996 and 2002 of the total expenditure on materials (matcost) minus the value of fuel and

electricity (energy), divided by the total value of shipments (vship). Post2002 is a dummy variable

equal to 1 every year after 2002, and 0 otherwise. The source of dependent variables is Pesquisa

Industrial Anual (PIA), years 1997-2009 (IBGE (2010)). The source of intermediate intensity is the

NBER Productivity database (Bartelsman and Gray (1996)). Robust standard errors clustered at

sector level are reported in parentheses. Significance levels: *** p<0.01, ** p<0.05, * p<0.1.

19

Table 7The effect of the reform on sectoral allocation of productionFalsification test of results reported in table 3: checking for pre-existing trends


Post1999×Intermediate Intensity 0.330 0.453 0.443 0.007

(0.213) (0.300) (0.299) (0.229)


Note: The table reports OLS estimates of the coefficient β in equation (1) in the text when the

regression is run on the years before the reform and the reform is supposed to happen in 1999. The

units of observation are 4-digits CNAE sector-year. Employment is the natural logarithm of the

total number of workers employed in establishments operating in year t and sector s. Revenues is

the natural logarithm of the total value of after tax revenues collected by establishments operating

in year t and sector s. Industrial sales is the natural logarithm of the after tax value of industrial

goods sold by establishments operating in year t and sector s, where the after tax value of sales is

estimated using the procedure suggested by IBGE (2010). Number of firms is the natural logarithm

of the total number of establishments operating in year t and sector s. Intermediate intensity is

calculated for every sector as the average between 1996 and 2002 of the total expenditure on

materials (matcost) minus the value of fuel and electricity (energy), divided by the total value of

shipments (vship). Post1999 is a dummy variable equal to 1 every year after 1999, and 0 otherwise.

The source of dependent variables is Pesquisa Industrial Anual (PIA), years 1997-2002 (IBGE

(2010)). The source of intermediate intensity is the NBER Productivity database (Bartelsman

and Gray (1996)). Robust standard errors clustered at sector level are reported in parentheses.

Significance levels: *** p<0.01, ** p<0.05, * p<0.1.

20

Table 8The effect of the reform on sectoral allocation of productionRobustness of results reported in table 3: the effect of the reform estimated

on 6 years of data


Post2002×Intermediate Intensity 0.548** 0.789** 0.810** 0.309

(0.232) (0.382) (0.374) (0.203)


Note: The table reports OLS estimates of the coefficient β in equation (1) in the text when

the regression is run on only 6 years (2000-2005). The units of observation are 4-digits CNAE

sector-year. Employment is the natural logarithm of the total number of workers employed in

establishments operating in year t and sector s. Revenues is the natural logarithm of the total value

of after tax revenues collected by establishments operating in year t and sector s. Industrial sales

is the natural logarithm of the after tax value of industrial goods sold by establishments operating

in year t and sector s, where the after tax value of sales is estimated using the procedure suggested

by IBGE (2010). Number of firms is the natural logarithm of the total number of establishments

operating in year t and sector s. Intermediate intensity is calculated for every sector as the average

between 1996 and 2002 of the total expenditure on materials (matcost) minus the value of fuel and

electricity (energy), divided by the total value of shipments (vship). Post2002 is a dummy variable

equal to 1 every year after 2002, and 0 otherwise. The source of dependent variables is Pesquisa

Industrial Anual (PIA), years 2000-2005 (IBGE (2010)). The source of intermediate intensity is the

NBER Productivity database (Bartelsman and Gray (1996)). Robust standard errors clustered at

sector level are reported in parentheses. Significance levels: *** p<0.01, ** p<0.05, * p<0.1.

21

taxes, misallocations and productivity · revenue and industrial sales. firms in the same sectors...

Documents